Google just released TPU v7 Ironwood, a chip that delivers 10X the performance of its predecessor. What looks like a search company is quietly becoming the infrastructure layer for the entire AI economy. Anthropic now runs its models on Google infrastructure. The Cloud backlog has nearly doubled. Waymo runs half a million paid rides every week. And while others wait for grid access, Google acquired a power infrastructure company and signed the first commercial deal for nuclear powered data centers. Google is building something larger than search. It is building the power grid, the chips, and the roads that will carry the next generation of technology.
What makes Google's AI chips different from Nvidia?
TPU v7 Ironwood delivers ten times the performance of its predecessor. But the real advantage is what these chips are built for. While Nvidia dominates training, Google optimized Ironwood for inference, the phase where AI models actually serve users. Nvidia owns the laboratory. Google is owning the factory floor. Analysis shows Ironwood costs roughly 44% less to operate than Nvidia at scale. Anthropic's commitment of a million TPUs signals a massive shift away from the GPU status quo.
How did Google solve the AI power crisis?
Google introduced a breakthrough in custom silicon with Ironwood. Now it is pushing the next wave of innovation in power. Every new data center needs grid connections that take years to approve. Google's solution was to buy the power directly. The Intersect acquisition integrates power generation directly into data center design, giving them behind the meter control. The Kairos nuclear deal puts reactors on their property. While competitors wait for utility approvals, Google generates its own electricity. Chips are in high demand. Power is the next constraint.
Why does Wall Street still undervalue Waymo?
Waymo reached 500,000 paid rides per week by December 2025. That translates to over a billion dollars in annual revenue run rate. With proven unit economics and a multi-year expansion runway, private funding rounds value it above a hundred billion dollars. Yet the stock still treats it as a research expense. Waymo is no longer burning cash for the future. It is generating revenue now. The science project became a business. And most analysts have not updated their models.
Is GOOGL stock expensive after a 65% rally?
The stock rallied 65% in 2025. On the surface, the multiple looks expensive. But the backlog tells a different story. Google Cloud's remaining obligations have nearly doubled year over year. That is locked in future revenue not yet on the income statement. When the backlog converts to revenue, the multiples compress. Add a new dividend and a seventy billion dollar buyback program. Google is now a dividend paying growth stock.
The last infrastructure shift this significant was the internet itself. Google won that era by organizing information. Now it is winning by owning the physical layer: the chips that run AI models, the data centers that house them, the nuclear plants that power them, and the vehicles that navigate the real world. The search bar is becoming invisible. The plumbing is becoming the product. Google's product is no longer answers. It is the infrastructure of the AI economy.